LightOn Open-Sources 149M Retrieval Models, Setting New BEIR SOTA
_reachsumit · x · 2026-07-30
LightOn has released a fully open-source recipe for training retrieval models, aiming to bridge the reproducibility gap caused by the reliance on closed training data in current SOTA models.
- Data & Models: Researchers reconstructed 665M English contrastive pre-training pairs from 34 public sources and built a fine-tuning dataset with mined hard negatives. This yielded two 149M-parameter models: DenseOn (single-vector dense) and LateOn (ColBERT-style late-interaction).
- Performance: These lightweight models achieve 56.20 and 57.22 average nDCG@10 on the BEIR benchmark, setting new state-of-the-art results for their size class.
- Multilingual Generalization: By translating the English data into 8 languages, the team trained 307M-parameter multilingual variants. Despite sharing a backbone, the dense model degrades on languages outside its translate-train support, whereas the late-interaction model generalizes better to unseen languages through token-level matching.
More from Research
- NeurIPS 2026 Creative AI Track Extends Deadline, Announces 'Agency' Theme — NeurIPSConf · 2026-07-31
- CoRT: Optimizing LLM Fine-Grained Credit Assignment via Counterfactual Replay — _akhaliq · 2026-07-31
- New MT3 Paradigm Enables Robots to Learn 1000 Tasks in Under 24 Hours — chris_j_paxton · 2026-07-31
- Peter Diamandis: We Are One AI-Discovered Material Away From Changing Civilization — PeterDiamandis · 2026-07-31
- Late Interaction Models Show Strong Generalization in Unseen Languages — antoine_chaffin · 2026-07-31
- Simulated LiDAR Needs Real-World Noise for Better Robustness — Sentdex · 2026-07-31